Parallel agents
Parallel subagent dispatch for independent work. Use when facing 2+ tasks that share no state and have no sequential dependencies — different test files, subsystems, or bugs that can be investigated concurrently.From its SKILL.md
npx -y skills add AllHailSeizure/synapse --skill parallel-agentsAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
2.6 KB, as published. Nobody here has run it
Parallel Agents
Dispatch one focused subagent per independent problem domain. They must not inherit session history — give each exactly the context it needs.
A subagent launch hook enforces this mechanically: briefs over 12,000 characters are refused as context forwarding, and more than two agents at once are refused until one finishes. Every seventh launch in a session asks you for a checkpoint. Send a task packet — task text, base commit, allowed files, done condition, one verify command — not session history. Workers are single-purpose and do not spawn further subagents; only you dispatch them.
Tune the caps per repo in .synapse/fanout.json
(maxConcurrent, maxPromptChars, waveSize).
Decision tree
- Multiple distinct problems?
- No → stay in this session.
- Independent (fixing one won't likely fix the others)?
- No → one agent (or you) investigates the related set.
- Can they run without shared mutable state (same files, same resources)?
- No → sequential agents or single session.
- Yes → parallel dispatch (multiple Task calls in one response).
Use when
- 3+ failing test files with different root causes
- Independent subsystems broken separately
- Each problem understandable without the others
Don't use when
- Failures are related
- Full-system context is required
- Exploratory debugging — you don't know what's broken yet
- Agents would edit the same files or fight over resources
Dispatch pattern
- Group by domain — one agent per domain.
- Write self-contained prompts — scope, goal, constraints, expected return.
- Fire in one response — multiple Task/subagent calls together = parallel.
- Integrate — read summaries, check conflicts, run the suite that covers the combined change.
Prompt checklist
- Focused: one file/subsystem, not "fix everything"
- Context: error messages, test names, relevant paths
- Constraints: what not to change
- Output: root cause + what changed
Good vs bad
| Bad | Good |
|---|---|
| "Fix all the tests" | "Fix agent-tool-abort.test.ts" |
| "Fix the race" (no where) | Paste failures + file path |
| No constraints | "Do not change production code" / "tests only" |
| "Fix it" | "Return root cause and diff summary" |
After return
- Review each summary
- Check overlapping edits
- Run verification covering the union of changes (
verificationskill)